What Google's Official AI Search Guide Means for E-Commerce: What to Keep Doing and What You Can Stop
Google has published an official guide titled "Optimizing your website for generative AI features on Google Search." In it, Google explains directly what you should do to appear in generative AI features such as AI Overviews and AI Mode.
So how should Shopify merchants respond to this guide? In this article, we've tried to organize the takeaways into three buckets: "what to keep doing / what you can stop doing / what to start working on."
The bottom line first: three actions
| Category | Specifics |
|---|---|
| ✅ Keep doing | Existing SEO best practices (content quality, technical SEO, keeping Merchant Center in order) |
| 🚫 OK to stop | Creating an llms.txt for Google Search, "chunking" your content, and rewriting it in an AI-friendly style |
| 🆕 Start doing | Following the Agentic Commerce Protocol (ACP) format that Shopify provides—setting accurate product categories, filling in category metafield values, and preparing AI-facing descriptions via Catalog Mapping |
There's an important caveat to "what you can stop doing." What Google called "unnecessary" applies only to Google Search. Supporting the Agentic Commerce Protocol (ACP) for ChatGPT, Copilot, and Perplexity is an effective measure for an e-commerce site. The fact that Shopify itself provides tools and endpoints for exactly this points to the same direction.
The passage that sparked the guide's attention
This guide drew attention because of the following sentence, which sent ripples through the industry.
"You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search."
(You don't need to create new machine-readable files, AI text files, markup, or Markdown to appear in generative AI search.)
The interpretation that "llms.txt is unnecessary" spread across social media, but the key point is that the subject of this sentence is Google Search.
Key points of Google's guide
How AI Overviews works — RAG and query fan-out
The guide states clearly that "there are no additional requirements to appear in AI Overviews or AI Mode; existing SEO best practices apply as-is." Meet Google Search's technical requirements and create content that helps people—the conventional approach is exactly what AI features evaluate as well. Behind this are two technical mechanisms the guide describes.
RAG (Retrieval-Augmented Generation) is a method that retrieves and references web pages from the core search ranking systems when generating an AI answer. Because it "builds AI answers on top of the search index," pages that aren't valued in the index won't show up in AI Overviews either.
Query fan-out is a mechanism that automatically generates multiple related subqueries from the user's original query. From a query like "lawn weed control," it processes derivative queries in parallel, such as "how to remove weeds without herbicide," "herbicides for lawns," and "how to prevent weeds."
In other words, the path to exposure in AI Overviews is an extension of organic search.
What you should do
What the guide emphasizes is continuing existing SEO best practices.
- Create non-commodity content: not content anyone could write, but content with a unique perspective grounded in firsthand experience and expertise
- Maintain technical SEO: pages that are crawlable, indexed, and allowed to show snippets
- High-quality images and video: because visuals are also cited in AI Overviews, image SEO best practices apply as-is
- Keep e-commerce and local business information in order: make use of the Merchant Center feed and Google Business Profile
What you don't need to do (mythbusting)
The guide also spells out what you don't need to do.
| Claimed to be "effective" | Google's view |
|---|---|
llms.txt creation |
Not needed |
| "Chunking" content (writing content split into small chunks so AI can process it more easily) |
Not needed |
| Rewriting style or structure for AI | Not needed |
| Gaming the system to gain mentions | Little effect (spam filters also catch it) |
| schema.org markup specific to AI Overviews | Not needed |
What "llms.txt is unnecessary" actually means
This is the single most important point of this article.
When Google declared llms.txt—along with the whole set of AI-oriented measures that have often come up in discussion—to be "unnecessary," that statement was limited to search visibility within Google Search.
Because Google uses RAG to reference indexed pages, AI Overviews works even without a special file like llms.txt. That much is correct.
But this is limited to Google Search. ChatGPT's shopping feature works by retrieving product data from the Shopify Catalog through the ACP (Agentic Commerce Protocol) defined by OpenAI. Google's statement has nothing to do with this axis.
| Technology | Google Search | ChatGPT / Copilot / Perplexity, etc. |
|---|---|---|
llms.txt |
Not needed (Google official) | Effective (used to retrieve a store overview) |
| ACP (Shopify Catalog) | Out of scope | Effective (the primary channel for delivering product feeds to ChatGPT) |
| schema.org Product | Effective (mainly for Rich Results) | Effective (crawled by GPTBot / ClaudeBot) |
Organizing e-commerce strategy for the AI era along two axes
Putting all of this together, it becomes clear that e-commerce strategy in the AI era needs to be considered along two independent axes.
| Axis | Goal | Evaluator | Main measures |
|---|---|---|---|
| ① Search visibility | Get found on Google | Google (including AI Overviews) | Continue existing SEO, non-commodity content |
| ② Agentic Commerce | Get AI to buy directly | ChatGPT / Copilot / Perplexity, etc. | Set up categories and metafields; prepare AI-facing descriptions as needed |
According to the ACP technical specification, ChatGPT's product selection is processed in two stages.
- Stage 1: Filtering by structured fields—candidates are narrowed down by category, price, inventory, and metafield values. If a product is dropped here, no matter how excellent its description is, it won't be read (the cutoff)
- Stage 2: NLP evaluation of the description—among the candidates that passed the filter, how well each matches the user's intent is evaluated via natural language processing (the scoring)
Metafields are the condition for passing the cutoff, while the description is the element that gets you selected—that's the relationship. Eliminating ambiguity is, you could say, the essence of preparing product data in the AI era.
What Shopify merchants should work on
Two moves that show how serious Shopify is
Around the same time as Google's guide, there were two notable moves on Shopify's side regarding e-commerce AI readiness.
One is the automatic deployment of AI Commerce endpoints across all stores. /llms.txt, /agents.md, and /.well-known/ucp now respond on every store without the merchant configuring anything. A path that lets AI execute payment directly—including Google Pay—is already open by default.
The other is the release of a verification tool called Agentic Readiness. It scores, across 11 items, how correctly an e-commerce site's product pages are recognized by AI agents—and what stands out is what it checks.
Shopify stores are out of scope for evaluation by this tool. The design assumes that if you're using Shopify, you already meet the requirements by default, so there's no need to check.
What these two facts show is the thoroughgoing stance Shopify is taking toward the era of AI agents. Shopify is trying to fully equip, at the infrastructure level, everything needed so that AI agents can accurately retrieve product information and so that no flaws arise in the AI-mediated purchase experience—preventing lost opportunities to the greatest extent possible.
When Google's guide says "no special measures for AI are needed," that is strictly about Google Search. Shopify's AI Commerce infrastructure—/llms.txt, /agents.md, and /.well-known/ucp—is aimed not at Google but at AI agents like ChatGPT, Copilot, and Perplexity. This is the "two axes" discussed earlier; the targets are fundamentally different. The correct perspective required in e-commerce operations in the AI era isn't "Google said so, so we don't need to do it"—it's correctly recognizing what each measure is aimed at, and then addressing both axes in parallel.
What merchants should actually do
"The platform builds the vessel, and the merchant fills it with data"—that's the structure of running a Shopify store in the AI era.
① Search visibility (the SEO axis)
- Stop writing commodity content and write original articles and product descriptions grounded in firsthand experience and expertise
- Maintain technical SEO (crawling, indexing, snippet settings)
- Keep the Merchant Center feed up to date
② The Agentic Commerce axis
- Set product categories accurately: choose the most specific node in the Shopify Standard Product Taxonomy. Once the category is set, the related metafields are auto-suggested
- Fill in category metafield values: structured attributes that AI uses for filtering, such as material, size, and care instructions. Empty fields don't match filter conditions, so the product can't pass the primary filter (the cutoff)
- Prepare separate AI-facing descriptions: by switching the source to a custom metafield in Catalog Mapping, you can set spec-focused text for AI distribution without changing the storefront description
-
Keep your store description in order: the beginning of
/llms.txtis auto-generated from this - Output reviews (aggregateRating) as structured data: check your Judge.me / Loox settings
That said, there are also cases where "we use Shopify, so we're safe" doesn't hold.
- Older themes (the Online Store 1.0 generation): the automatic output of schema.org may be incomplete.
- Custom themes that override JSON-LD: the structured data may be broken unintentionally.
Summary
What Google's official guide presented is a simple message: "No special measures for AI search are needed—keep doing your existing SEO." But this is limited to Google Search.
It's more accurate to organize e-commerce strategy in the AI era along two axes.
- Search visibility (Google): continue existing SEO. Non-commodity content is the differentiator
- Agentic Commerce (ChatGPT, etc.): fill in values for categories and metafields, and prepare AI-facing descriptions. The ACP-ready vessel is already provided automatically by Shopify
"Eliminate ambiguity and increase the amount of information AI can interpret accurately"—this is the essence of the work shared by both axes.